Spotting fake Trustpilot reviews

I made VPS last month at new provider, very happy with service, then I check Trustpilot and see 47 reviews in one week all saying "excellent uptime" and "fast support." Made me suspicious, so I analyze little bit.

Patterns I find in fake reviews:
- same phrases repeated: "blazing fast," "never had issue"
- reviewer has only one review ever
- all posted in short window
- username is Firstname lastname with number
- no mention of specific problem or use case

I collect 12 reviews from 3 providers, can share if useful. Thank you very much sir for any feedback on my method, I am still learning.

Your methodology needs refinement. Specific issues:

1. Temporal clustering alone is insufficient—legitimate launch reviews cluster too.
2. Single-review accounts are common for first-time VPS buyers who found you via search.
3. "Firstname lastname" format matches Trustpilot's default suggestion.

Better indicators:
- Identical sentence structures across accounts (parse tree similarity >0.85)
- Reviewer location mismatches with provider's actual customer base
- Response time claims that don't match published SLA
- Absence of version-specific references (e.g., "Debian 12 template had pre-installed fail2ban")

Practical recommendation: Cross-reference with Web Archive for review timing vs. Provider age. A 5-year-old provider with 200 reviews all from last 3 months is suspect.

I scraped 500 reviews with a simple Python script, fed them into a local LLM via Ollama, asked it to flag astroturfing patterns. Runs on a $4 HostHatch instance with docker compose.

The real insight isn't the fake reviews—it's that most "independent" hosting comparison sites are just affiliate farms. Self-host your research. Reverse proxy the whole stack behind nginx, job done.

I can share the compose file if anyone wants to replicate.